Two experiments investigated the mental representation of spatial descriptions. In Experiment 1, the subjects classified a series of diagrams, each presented after a spatial description, as either consistent or inconsistent with the description. They were then given an unexpected recognition test of their memory for the descriptions. The subjects remembered the meanings of determinate descriptions very much better than those of grossly indeterminate descriptions; their memory for a description was not reliably affected by whether or not the diagram had been consistent with it. Experiment 2 extended these findings and showed that, although the semantic implications of a determinate description are better remembered than are those of an indeterminate description, the verbatim details of an indeterminate description are easier to recall than are those of a determinate description. The results are taken to imply the existence of two different sorts of encoding: propositional representations that are relatively hard to remember but correspond closely to the sentences in the description, and mental models that are relatively easy to remember but are analogous to spatial arrays and accordingly poor in linguistic detail.